Professional Documents
Culture Documents
www.elsevier.com/locate/ijhmt
a
Duke University, Department of Mechanical Engineering and Materials Science, Durham, NC 27708-0300, USA
b
FAST UMR 7608 (CNRS Univ. Paris VI and XI), Bat. 502, 91405 Orsay, France
Abstract
In this paper we rely on the constructal law of maximization of ow access in order to construct a theory of geometry generation
(selection, evolution) during molten droplet impact. We show that immediately after impact the liquid spreads inviscidly as a ring with
a radial velocity that scales with the initial impact velocity. If the initial droplet is small and slow enough, the splat comes to rest (dies)
viscously, as a disc. If the droplet is large and fast enough, the ring splashes and is continued outward by needles that grow radially until
they are arrested by viscous eects. We optimized the number of needles such that the total splash time is minimum. The theoretical
dimensionless group that governs the selection of geometry (G) is the ratio of two lengths, the nal radius of the disc that dies viscously,
divided by the radius of the still inviscid ring that just wrinkles. Splats form when G 6 O(1) and splashes are favored when G P O(1).
Experimental measurements reported in the literature conrm several of the features of the constructal development of splat vs. splash
ow architecture.
2006 Elsevier Ltd. All rights reserved.
Keywords: Constructal theory; Geometry generation; Selection; Evolution; Molten droplet; Splashing; Splat; Spray coating; Fingering; Forensic medicine;
Blood splatter
0017-9310/$ - see front matter 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijheatmasstransfer.2006.02.001
A. Bejan, D. Gobin / International Journal of Heat and Mass Transfer 49 (2006) 24122419 2413
Nomenclature
where the half-domain of splashing behavior is occupied by accelerates the arrival of equilibrium. The predictive pro-
OhmRep > constant [2]. More recently, Mundo et al. [3] gress made by invoking the constructal law is reviewed in
conducted an extensive experimental program that led to [6,7], and is not reviewed again here. In this paper, we show
a correlation of the same type, We1/2Re0.25 = K (which is that everythingthe time and length scales of the liquid
the same as Oh Re1.25 = K), where the constant is K = ow before and after splashingresults from the invoca-
57.7 for a set of experiments using water, ethanol or tion of this principle, including the power-law boundary
water-sucrose-ethanol droplets impacting a smooth stain- We (Re) and its two constants K and 0.25 (the theoretical
less steel substrate. More experimental data are provided value of the latter turns out to be 1/4, analytically). Most
by two recent theses [4,5]. remarkable is the fact that the analysis reported in this
Correlations that are tight, robust and conrmed by new paper was formulated not to derive Mundo et al.s cor-
experiments call for a theory to explain their origin, and to relation (the discovery of the correlation was one of the
predict them. Mundo et al. [3] proposed such a theory, surprises of this work), rather the objective was to see
which was based on a global accounting for the conserva- whether the principle of maximization of ow access antic-
tion of liquid energy and mass during droplet impact. Their ipates the splashing phenomenon as successfully as it did
analysis generated an expression for the boundary between other natural ow architectures such as turbulent structure,
splashing and no splashing, which next to Oh and Re con- Benard convection and dendritic solidication [6,7], and,
tained two additional parameters: the contact angle and the more recently, the architecture of the human lung [8], the
dimensionless ratio Dmax/D, where D is the original droplet hydraulic jump phenomenon [9] and the speed and fre-
diameter and Dmax is the maximal diameter of a liquid quencies (stride, apping) of all ying, running and swim-
spreading as a disc. Mundo et al.s formula passes through ming animals [10].
the Oh(Re) data that mark the splashing/no splashing From the outset, we admit that the complexity of the
boundary, but deviates from the data when Re < 200 and splashing droplet phenomenon forces us to formulate the
Re > 2000. Furthermore, the Mundo et al. formula does discussion in terms of scale analysis [11]. Furthermore,
not have the power-law structure of the empirical correla- complexity makes it necessary to dene unambiguously
tion We1/2Re0.25 = K, in which one should note that there the various scales of the phenomenon, which change from
are two empirical constants that were determined by opti- time to time. We have done this carefully in the following
mizing the curve tting of the data: K and the exponent analysis, and we draw the readers attention to the deni-
0.25. tion of the subsequent time scales (t1, t2, t3) and on the sce-
In this paper we report a purely theoretical argument nario that emerges.
that anticipates not only the observed transition between
splashing and no splashing, but traces the entire scenario 2. Scale analysis of ring-shaped ow
of droplet impact behavior to the action of the constructal
law [6,7]: the ow geometry that is selected evolves in time Consider a spherical liquid droplet of diameter D, vis-
such that it maximizes the spreading of the liquid, and cosity l and density q, which impacts a solid wall. The
2414 A. Bejan, D. Gobin / International Journal of Heat and Mass Transfer 49 (2006) 24122419
r Vt 5
This means that the splat radius moves outward with a
speed of the same order as the speed of impact,
r
u V 6
t
Let S be the cross-sectional area of the splat (in the
plane of Fig. 2) at the time t, such that the conservation
of droplet liquid volume requires
D3 rS 7
At this stage we do not know whether the splat is a at disc
or a ring: the shape of the splat will be determined next.
The linear momentum of the droplet before impact is
Fig. 1. Droplet parameters before and during spreading on the wall.
(qD3)V. The linear momentum of the splat liquid is (qSr)Ua
where Ua is the S-averaged radial velocity with which the
splat liquid moves. The conservation of liquid linear
droplet velocity V is perpendicular to the wall at r = 0, as
momentum requires:
shown in Fig. 1. The liquid droplet becomes a splat that
spreads on the wall. We are interested in the scales (length, D3 V rSU a . 8
time) and the morphology of the splat. This means that the Eqs. (7) and (8) require
following scenario refers to times longer than the initial
time interval of splat deformation (t D/V), during which Ua V 9
the droplet touches the wall and is deformed. According to Eq. (6), V is the velocity of the splat radius.
Let t = 0 mark the moment when the splat is born on Therefore, Eq. (9) states that all the splat liquid moves
the wall. This movement is indicated by the deformed radially with the same speed as the splat radius, and from
droplet shown in the lower part of Fig. 1. The stagnation this follows the conclusion that the splat liquid must be
of the liquid causes an instantaneous excess pressure on located close to its rim. This is why in Fig. 2 the splat liquid
the order of is shown as a ring of radius r and cross-section S.
1 The length scale of the ring cross-section is S1/2. This
P qV 2 1 statement is equivalent to the assumption that the contact
2
angle between the liquidair surface and the solid surface is
This pressure drives the splat liquid to the distance r, which
not innitesimally small, i.e. that the shape of the ring
is the instantaneous radius of the splat. We assume that
cross-section S is not slender. This assumption covers most
immediately after t = 0 the splat is symmetric about the
of the known liquidairsolid contacts, which in the pres-
r = 0 axis, and its ow is inviscid. The momentum equation
ent scale analysis are represented by cross-sectional shapes
in the r direction is
(S) that have a single length scale (S1/2), not two, a width
ou ou oP
q u 2
ot or or
where u is the liquid velocity component in the r direction.
From kinematics, the radial velocity scale is
r
u 3
t
According to the method of scale analysis [11] the momen-
tum Eq. (2) expresses the competition between three scales:
u u P
q ; qu ; . 20
t r r
In view of Eq. (3), the rst two scales are both of order
qr/t2. In conclusion, the momentum Eq. (2) requires the
following balance of two scales:
r P
q 4
t2 r
Combining Eqs. (4) and (1), and neglecting (at this stage)
factors of order 1, we conclude that the splat radius in-
creases linearly in time, Fig. 2. Ring-shaped spreading of the liquid.
A. Bejan, D. Gobin / International Journal of Heat and Mass Transfer 49 (2006) 24122419 2415
and a thickness. Next, by combining Eqs. (5) and (7) we Here 10 is the order of magnitude of the factor (2p31/2 or
nd that the ring becomes thinner in time, 2p61/2) that appears in the exact solution to the stability
problem [12]. In place of qg, the force that pushes the splat
D3
S 10 liquid against the r-radius free surface is P/r qr/t2
tV qV/t. Eq. (16) becomes
Friction works toward slowing the spreading of the 1=2
splat. The thickness of viscous penetration from the wall rt
k 10 17
transversally through the splat is qV
d mt
1=2
11 The wavelength of rim instability increases as t1/2, whereas
the rim radius increases as t, as shown in Fig. 4. At short
where m is the liquid kinematic viscosity. The splat ow be- times, r is smaller than k, and the splat remains round.
comes viscid, and its spreading stops when viscous diu- At suciently long times, the rim is long and at enough
sion sweeps the entire splat cross-section. This happens to accommodate a ripple of several (n) wavelengths,
when d S1/2, which in view of Eqs. (10) and (11) becomes
the time scale r nk 18
3 1=2 where n is a factor larger than 1, but not much larger, for
D
t1 12 example O(10). The time t2 when Eq. (18) becomes true is
mV estimated from Eqs. (17) and (18)
The splat radius at this time is r
t2 10n2 19
3 1=2 qV 3
VD
r1 13 When t2 < t1, the splat develops a wavy rim before its
m
motion is arrested by wall friction. This inequality is anal-
or ogous to the comparison of two time scales that was used
r1 to predict ow morphology in Benard convection, lami-
Re1=2 14
D nar-turbulent transition, and dendritic crystal growth
where Re is the Reynolds number of the original droplet, [6,7]. By using Eqs. (12) and (19), we nd that the splashing
condition t2 < t1 can be written as
VD
Re 15
m We1=2 Re1=4 > 10n K 20
In conclusion, Eq. (14) shows that larger and faster drop- where We is the droplet Weber number,
lets make relatively larger splats.
qDV 2
We 21
3. Splashing and needle-shaped ow r
The splat ring radius at the time t = t2 is
Next, we turn our attention to the eect of surface ten- r2 10n
2
Fig. 4. The growth of the ring radius and the wavelength of Taylor
Fig. 3. The wavelength of Taylor instability. instability.
2416 A. Bejan, D. Gobin / International Journal of Heat and Mass Transfer 49 (2006) 24122419
Fig. 5. The liquid enters with the velocity V through the Next, we make two additional observations in line with
r = r2 end of the nger. Its bottom layers are slowed down the constructal law of maximization of ow access, or
viscously by the wall, while fresh liquid proceeds (over the acceleration of approach to equilibrium. First, the total
top) to longer radial distances. splash time (t2 + t3) can be minimized by selecting the num-
The farthest reach (L) can be estimated in two ways, as ber of ngers. The optimal number is
follows. Let A be the scale of the nger cross-sectional area, 2=5
and A1/2 the length scale of the nger cross-section. This is We Re1=2
nopt 29
again the wettability assumption invoked above Eq. (10), 4 102
according to which contact angles are not innitesimally with the corresponding minimum splash time
small, and the liquid cross-sectional shape has a single 1=5
length scale (A1/2), not two. The nger control volume is t2 t3 min 5 102 r2 Lmin
4=5 30
pushed from the left by the impulse qV2A, and from the t1 4 We Re 1=2 r1
right by the wall friction force,
These results are order-of-magnitude accurate because they
qV 2 A sLA1=2 23 come from expressions (27), (28), which are based on scale
where s lV/AS1/2. In conclusion, the nal length of the analysis. Eqs. (29) and (30) can also be obtained in an
nger is order-of-magnitude sense by setting the two terms of Eq.
(27) equal to each other. This alternate approach is the
AV intersection of asymptotes method, which has been used sev-
L 24
m eral times in constructal theory to predict natural ow con-
Mass conservation requires nSL D3, and, after using gurations, e.g., the rst eddy of a turbulent ow, the onset
Eq. (24), the dimensionless nger length becomes of Benard convection, the birth of needles in dendritic crys-
1=2 tals and the characteristic length scale of cracks in volumet-
L Re rically shrinking solids [6,7].
25
D n The second observation is that in view of the splash-
formation criterion (20), the splash time is shorter than
The time interval during which the nger grows is
the ring time,
L A
t3 26 t2 t3 min < t1 31
V m
If ngers can form, they will form, because ngers promise
This t3 expression suggests an alternative calculation of L: to bring the liquid to rest faster than the ring-shaped ow.
the nger length is the distance traveled by the V liquid This time minimization criterion is the same as the con-
during the time required by that liquid vein to be pene- structal law of maximization of ow access [6,7].
trated by viscous diusion through its thickness A1/2. That
time is such that A1/2 (mt3)1/2, which leads to Eq. (26) 4. Dimensionless summary of ring and needle ows
and, after invoking mass conservation, to Eq. (25).
To summarize, we determined the nal length and time One of the attributes of theory is that it organizes the
scales of a splat that remains ring shaped until its viscous presentation of the phenomenon in terms of the smallest
death (r1, t1). We also determined the nal length scale and most relevant dimensionless groups. In this paper we
(r2 + L) and total time scale (t2 + t3) of a round splat that have identied several length and time scales. If the post-
splashes into n ngers. Taking the (r1, t1) scales as refer- impact ow of liquid is ring shaped (Fig. 2), then the time
ence, we nd that the ring and splash scales are of the same of the length and time scales when the growth of the ring is
order of magnitude (see Eq. (20)): stopped by viscous diusion are
2
r2 L 10n 1 r1 t1 m
1=2 1 27 Re1=2 Re1=2 32
r1 We Re 1=2 n D D2
2
t2 t3 10n 1 The time when the ring develops periodic deformations and
1=2
1=2 1 28
t1 We Re n ngers start to grow is characterized by the scales
A. Bejan, D. Gobin / International Journal of Heat and Mass Transfer 49 (2006) 24122419 2417
2 2
r2 10n t2 m 10n and (41), is a consequence of the constant velocity scale
33 of the ring and nger uid, V.
D We D2 ReWe
Dividing Eq. (33) by Eq. (32), we nd how the scales of the
5. Experimental conrmation
beginning of splashing (r2, t2) are smaller than the scales of
the ring-shaped droplet at the end of its symmetric growth
The scaling laws uncovered in this paper are now com-
(r1, t1):
pared with a set of experimental results published recently
2
r2 t2 10n [5,13], where the experimental device was designed to
<1 34 record the shape evolution of droplets falling by gravity
r1 t1 We Re1=2
on solid substrates. Results are presented for millimetric
These conclusions can be rened further by using the nopt droplets (13 mm in diameter) with relatively low velocities
scale of Eq. (29) and introducing it in Eqs. (20) and (34). (14 m/s) and substrates of dierent roughness and wetta-
Accordingly, K 10n is bility. The range of corresponding Reynolds numbers
10 extends from 3500 to 12500, which corresponds to Weber
K We2=5 Re1=5 35 numbers in the range 57254.
4002=5
The rst theoretical result that we compare with exper-
Introducing this K estimate into the splashing criterion iments is the order of magnitude of the radial velocity Ua
t2 < t1, or Eq. (20), we obtain after impact. The analysis [Eq. (9)] shows that the initial
1=5 10 radial velocity is on the order of the impact velocity. The
We1=2 Re1=4 > 2=5
1 36 experimental data available in Ref. [5] and displayed in
400
Table 1 are for two impact velocities, V = 1.2 and 3.6
We may compare this theoretical criterion with the experi- m/s. The tabulated data show that for both sets of experi-
mental correlation [3] ments the ratio Ua/V is nearly constant. This means that
1=5 Eq. (9) is veried, with a proportionality factor of 3.5 (with
We1=2 Re0:25 > K 1=5 37
standard deviation of 10%), which is a factor on the order
1/5
to see that K = 57.7 1, which means that the theoreti- of 1.
cal criterion (36) agrees remarkably well with experimental Another important feature of the experimental data is
data. To this comparison we will return in Table 2. that the ratio Ua/V is independent of the nature of the
Another eect of the optimized number of needles impact surface. This is in accord with the analysis that
becomes visible when we substitute Eq. (20) into Eq. (34): led to Eq. (9), which is based on the observation that the
initial ow is inviscid, hence unrelated to the wall
r2 t2 100=4004=5 1 condition.
38
r1 t1 We Re1=2 1=5 G A systematic observation of the splat structure evolution
was conducted in Ref. [5], by increasing step by step the
where impact velocity of droplets of the same diameter. The aver-
G We Re1=2
1=5
39 age value of the highest velocity for spreading and the low-
1/2
est velocity for splashing yields a transition velocity, with
Unlike WeRe , the new dimensionless group G has a a typical accuracy of 5%. Several substrates (glass, wax,
physical meaning: G is the ratio of two lengths PVC) and roughnesses ranging from 0.003 to 60 lm were
final radius of disc that dies viscously used in the experiments, and this led to signicant scatter
G 40 in the results.
radius of still inviscid ring that just wrinkles
The values of the corresponding critical values for
This is why G P O(1) is the criterion for the occurrence of We1/2Re1/4 are shown in Table 2. Recall that We and Re
wrinkles on the rim at time t2: the wrinkles grow as needles are based on liquid properties and initial droplet diameter
from t2 to t2 + t3. On the other hand, when G 6 O(1), the and velocity. In spite of the large scatter in the results, this
splat stops owing when it is round, and consequently it table clearly shows the relevance of the group We1/2Re1/4 as
has no future as a shape other than the round disc at rest.
Finally, to compare the scales of the splash just before Table 1
vs. end of splashing, we divide Eq. (30) by Eq. (38): Experimental measurements of the initial spreading velocity Ua: the
tabulated values are for the ratio Ua/V (data from [5])
r2 Lmin t2 t3 min
5 41 Substrate V = 1.2 m/s V = 3.6 m/s
r2 t2
Smooth wax 3.42 3.30
These ratios of length and time scales are the same constant Smooth glass 3.45 3.21
which is on the order of 1. The size of the splashed droplet Rough wax 3.98 3.49
size is proportional to the size that the splat has at the start Rough glass 3.84 3.21
of asymmetric deformations. The fact that the ratios of PVC 300 3.41 3.69
PVC 100 3.38 3.63
length scales match the ratios of time scales, Eqs. (30)
2418 A. Bejan, D. Gobin / International Journal of Heat and Mass Transfer 49 (2006) 24122419
(compact, hierarchical, constructal) can our brains main- [6] A. Bejan, Advanced Engineering Thermodynamics, 2nd ed., Wiley,
tain the outow (science, knowledge), in spite of our New York, 1997 (Chapter 13).
[7] A. Bejan, Shape and Structure, from Engineering to Nature,
nite-size anything (cranium, food, life time). Cambridge University Press, Cambridge, UK, 2000.
[8] A.H. Reis, A.F. Miguel, M. Aydin, Constructal theory of ow
Acknowledgement architecture of the lungs, Med. Phys. 31 (5) (2004) 11351140.
[9] A. Pramanick, P.K. Das, Note on constructal theory of organization
The authors wish to thank the Universities of Paris VI in nature, Int. J. Heat Mass Transfer 48 (2005) 19741981.
[10] A. Bejan, J.H. Marden, Unifying constructal theory for scale eects in
and Paris XI for inviting AB as a visiting professor at the running, swimming and ying animals, J. Exp. Biol. 209 (2006) 238
FAST Laboratory. 248.
[11] A. Bejan, Convection Heat Transfer, third ed., Wiley, New York,
References 2004, p. 454.
[12] J.H. Lienhard, A Heat Transfer Textbook, Prentice-Hall, Englewood
[1] S. Haferl, D. Attinger, Z. Zhao, J. Giannakouros, D. Poulikakos, Clis, NJ, 1981, pp. 399400.
Transport phenomena in the impact of a molten droplet on a surface: [13] R. Rioboo, C. Bauthier, J. Conti, M. Voue, J. De Coninck,
macroscopic phenomenology and microscopic considerations. Part I: Experimental investigation of splash and crown formation during
Fluid dynamics, Annu. Rev. Heat Transfer 11 (2000) 65143. single drop impact on wetted surfaces, Exp. Fluids 35 (6) (2003) 648
[2] C.D. Stow, M.G. Hadeld, An experimental investigation of uid 652.
ow resulting from the impact of a water drop with an unyielding dry [14] Z.-Z. Xia, Z.-X. Li, Z.-Y. Guo, Heat conduction optimization: high
surface, Proc. Roy. Soc. London A 373 (1981) 419441. conductivity constructs based on the principle of biological evolution,
[3] Chr. Mundo, M. Sommerfeld, C. Tropea, Droplet-wall collisions: in: 12th International Heat Transfer Conference, Grenoble, France,
experimental studies of the deformation and breakup process, Int. J. 1823 August 2002, vol. 2, pp. 2732.
Multiphase Flow 21 (1995) 151173. [15] T. Borvall, A. Klarbring, J. Petersson, B. Tostenfelt, Topology
[4] C. Escure, Etude de limpact des gouttes dalumine sur une cible optimization in uid mechanics, in: H.A. Mang et al. (Eds.), Fifth
chaude ou un lm dalumine liquide, Doctoral Thesis, University of World Congress on Computational Mechanics, Vienna, Austria, 712
Limoges, France, 2000. July 2002.
[5] R. Rioboo, Impact de gouttelettes sur surfaces solides et seches, [16] A. Bejan, S. Lorente, The constructal law and the thermodynamics of
Doctoral Thesis, University Paris VI, 2001. Available from: <http:// ow systems with conguration, Int. J. Heat Mass Transfer 47 (2004)
ftp.espci.fr/incoming/goutte/These_Rioboo.pdf>. 32033214.